Skip to content
hp.Hunain ParekhAI process automation engineerLet’s talk

Four processes.
No one in the loop.

Each one used to need a person at every step. The green line on every case says where a human is still involved, and the numbers say how much runs without one.

BeMeSystem illustration
Can we schedule a visit for Friday?
Business context retrievedChecking calendar availability
Let’s find a time that works.
Booked. No human involved.

Sales conversations that close themselves.

AI assistance across messaging, email, and voice. Shared business knowledge meets tools that schedule appointments and keep conversations moving.

A person steps in only when the customer asks for one or identity cannot be confirmed. Everything else, including the booking, runs without a human.

20+ business owners onboarded. 1M+ calls handled.

Agent toolsKnowledge retrievalCross-channel context

How it works

WhatsApp, Instagram, Messenger, SMS, Gmail, and Outlook workers connect to a shared AI response system. Context providers combine business information, conversation history, and document retrieval. Channel-specific prompts shape the response.

Engineering detail

Customer identity confirmation controls access to cross-channel history. Appointment tools connect to Google and Microsoft calendars and check availability before booking.

xAIaSystem illustration
Call in. Appointment out.
ArabicEnglishSimplex

Patient booking, from the call to the record.

Voice and WhatsApp agents that book patient appointments directly into the Simplex system, and push leads into the CRM without anyone retyping them.

The appointment lands in Simplex during the call. If the CRM sync fails, it queues and replays. A human sees it only if the replay fails too.

30k+ calls and 350k conversations a month. 63 live agents in production.

Voice agentsMulti-turn retrievalCRM integrations

How it works

Organization-scoped conversations, a bounded knowledge-retrieval loop, and queued lead processing connect the AI experience to business operations. The Odoo integration includes a failed-sync queue and replay controls.

Engineering detail

A custom Hamsa–Vapi bridge selects Arabic or English voices, supports Arabic dialect configuration, and converts synthesized WAV audio into PCM at the requested sample rate.

Play LoungeSystem illustration
You’re on the list.Good times.
Great company.
Play Lounge
Guest experience

Guest admission with no one at the list.

An event platform connecting venues, host requests, invitations, RSVPs, and guest admission in one operational flow.

Invitations, RSVPs, and QR admission run on their own. Staff handle only the exceptions the system flags, and check-in state survives the list changing mid-event.

1k+ venue bookings a month.

Event operationsQR check-inAccess control

How it works

Venue and event management connect to guest invitations, RSVP responses, QR admission, staff coordination, and automated email communication.

Engineering detail

Guest-list updates preserve check-in state during an event. Role-based access and dedicated verification-team workflows support the people managing admission.

Track LoopSystem illustration
TaskInvoice module shippedSubmitted2 days earlyQAPassed, 1 minorLive feedbackClient: “exactly what we asked”
Performance this week92

Performance measured from the work, not a form.

Time tracking software where the AI scores employee performance from the task itself: what was done, when it was submitted, and how QA and live feedback rated the quality.

No review forms. The score builds continuously from every task. A manager looks only when a score moves sharply or feedback and QA disagree.

Prototype. Every feature above was built and demonstrated end to end; it has not run inside a company yet.

Task trackingQA scoringLive feedback

How it works

Tasks, submission times, QA results, and live feedback flow into one record per person. The model weighs on-time delivery against quality signals and produces a running performance view instead of a quarterly opinion.

Engineering detail

Quality is judged from QA outcomes and feedback on the actual submission, never from hours logged alone. The exceptions list surfaces sharp changes and conflicting signals for a human to read.

Code you can read.

Two repositories that show how I build,
outside client work I cannot publish.

Maqsam to Odoo syncCall records become CRM leads on a schedule. Deduplication, paging, and a state file that only advances after every record lands, so a failed Odoo run retries the same window instead of losing calls.Node.js, n8n, OdooRepoGuardA git-aware scanner for injected code, malicious dependency changes, committed secrets, and history rewrites. Scores the signals a payload cannot drop and still work.Go

What does your team
still do by hand?

Tell me about it